`act_runner`'s built-in GitHub-Actions-cache server binds inside the
runner container on the compose-defined `apf-portal-act-runners`
bridge. Jobs spawned via the mounted Docker socket come up on
Docker's default `bridge` network and cannot reach it. Every job
opting into `cache: 'pnpm'` therefore eats ~2 min ETIMEDOUT on
restore at the start and another ~2 min on save at the end — for
zero cache hits. Across the five jobs that's ~20 min wasted per CI
run.
Drop `cache: 'pnpm'` from all five jobs. `pnpm install
--frozen-lockfile` is fast enough on the warm store inside the job
container that the cache layer is currently no value-add anyway.
Document the proper fix (cross-container networking / fixed-port
cache binding) in infra/README.md so a future infra spike can
re-enable caching cleanly.
Julien Gautier